Mistral has released its latest open-weight frontier model, Mistral Large 4 (ML4), also known as le Chonk, a 1 trillion-parameter offering that it says competes with rivals from China.

ML4 is proficient at security, programming, manufacturing, finance and multimodal tasks, Mistral said.

It indicated the model, which is currently being provided to a limited group of customers, would rank among the top global open-weight models on aggregate benchmark performance, and was the strongest open-weight model developed outside China by a “substantial margin”.

“The model capabilities will further improve as we scale up our training capacity, following our Series D fundraise,” said Guillaume Lample, Mistral co-founder and chief scientist.

“Further, the cyber defence capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models to perform cyber attacks.”

Open-weight competition

Chinese companies including Alibaba, DeepSeek, Moonshot and Z.ai have this year offered open-weight models that narrow the gap with closed-weight offerings from OpenAI and Anthropic.

In the West, the market for open-weight models is less developed, but this week Brooklyn-based start-up Reflection AI released its first frontier offering, the open-weight model Beam, which it said matched the performance of Z.ai’s GLM-5.2, the predecessor to GLM-5.2, released in August.